Support GZIP encoded POSTs/GETs via CURL

[staabm]

we use the API and send between 15k and 40k requests per hour.. things get pretty slow..

we aleady batch the requests to reduce the overhead.

after a bit of investigation it looks like requests/responses are not compressed/gzip-encoded. we only see Accept-Encoding: deflate, gzip but no Content-Encoding: gzip

also when posting items to the webservice it seems things went across the wire in plain text.

see the following output, after hacking curl_setopt($this->handle, CURLOPT_VERBOSE, 1); into the cURL.php of the http-client:

*   Trying 136.243.64.171...
* Connected to rapi.recombee.com (136.243.64.171) port 443 (#0)
* ALPN, offering http/1.1
* Cipher selection: ALL:!EXPORT:!EXPORT40:!EXPORT56:!aNULL:!LOW:!RC4:@STRENGTH
* successfully set certificate verify locations:
*   CAfile: /cluster/www/www/www/telexa/vendor/rmccue/requests/library/Requests/Transport/cacert.pem
  CApath: /etc/ssl/certs
* SSL connection using TLSv1.2 / ECDHE-RSA-AES256-GCM-SHA384
* ALPN, server accepted to use http/1.1
* Server certificate:
*        subject: OU=Domain Control Validated; CN=*.recombee.com
*        start date: Dec 30 10:26:13 2018 GMT
*        expire date: Feb 28 16:23:00 2021 GMT
*        subjectAltName: rapi.recombee.com matched
*        issuer: C=US; ST=Arizona; L=Scottsdale; O=GoDaddy.com, Inc.; OU=http://certs.godaddy.com/repository/; CN=Go Daddy Secure Certificate Authority - G2
*        SSL certificate verify ok.
> GET /telexa-dev/items/list/?hmac_timestamp=1585732533&hmac_sign=XXXHTTP/1.1
Host: rapi.recombee.com
Accept: */*
Accept-Encoding: deflate, gzip
Referer: https://rapi.recombee.com/telexa-dev/items/list/?hmac_timestamp=1585732533&hmac_sign=XXX
User-Agent: recombee-php-api-client/3.0.0
Connection: close

< HTTP/1.1 200 OK
< Server: nginx/1.14.1
< Date: Wed, 01 Apr 2020 09:15:33 GMT
< Content-Type: application/json; charset=utf-8
< Content-Length: 190555
< Connection: close
< X-Recombee-Request-Id: 2100e32cfc2677b0f662f571fccd69ea
< Cache-Control: no-cache
<
* Closing connection 0
DIE!mstaab@mst16:/cluster/www/www/www/telexa/scripts$ php recombee.php partner development 17
start:2020-04-01 11:16:37
*   Trying 136.243.64.171...
* Connected to rapi.recombee.com (136.243.64.171) port 443 (#0)
* ALPN, offering http/1.1
* Cipher selection: ALL:!EXPORT:!EXPORT40:!EXPORT56:!aNULL:!LOW:!RC4:@STRENGTH
* successfully set certificate verify locations:
*   CAfile: /cluster/www/www/www/telexa/vendor/rmccue/requests/library/Requests/Transport/cacert.pem
  CApath: /etc/ssl/certs
* SSL connection using TLSv1.2 / ECDHE-RSA-AES256-GCM-SHA384
* ALPN, server accepted to use http/1.1
* Server certificate:
*        subject: OU=Domain Control Validated; CN=*.recombee.com
*        start date: Dec 30 10:26:13 2018 GMT
*        expire date: Feb 28 16:23:00 2021 GMT
*        subjectAltName: rapi.recombee.com matched
*        issuer: C=US; ST=Arizona; L=Scottsdale; O=GoDaddy.com, Inc.; OU=http://certs.godaddy.com/repository/; CN=Go Daddy Secure Certificate Authority - G2
*        SSL certificate verify ok.
> GET /telexa-dev/items/list/?hmac_timestamp=1585732597&hmac_sign=XXXHTTP/1.1
Host: rapi.recombee.com
Accept: */*
Accept-Encoding: deflate, gzip
Referer: https://rapi.recombee.com/telexa-dev/items/list/?hmac_timestamp=1585732597&hmac_sign=XXX
User-Agent: recombee-php-api-client/3.0.0
Connection: close

< HTTP/1.1 200 OK
< Server: nginx/1.14.1
< Date: Wed, 01 Apr 2020 09:16:37 GMT
< Content-Type: application/json; charset=utf-8
< Content-Length: 190555
< Connection: close
< X-Recombee-Request-Id: 06730ddebc72699bdc2f513d5b3241d0
< Cache-Control: no-cache
<
* Closing connection 0
* Hostname rapi.recombee.com was found in DNS cache
*   Trying 136.243.64.171...
* Connected to rapi.recombee.com (136.243.64.171) port 443 (#0)
* ALPN, offering http/1.1
* Cipher selection: ALL:!EXPORT:!EXPORT40:!EXPORT56:!aNULL:!LOW:!RC4:@STRENGTH
* successfully set certificate verify locations:
*   CAfile: /cluster/www/www/www/telexa/vendor/rmccue/requests/library/Requests/Transport/cacert.pem
  CApath: /etc/ssl/certs
* SSL connection using TLSv1.2 / ECDHE-RSA-AES256-GCM-SHA384
* ALPN, server accepted to use http/1.1
* Server certificate:
*        subject: OU=Domain Control Validated; CN=*.recombee.com
*        start date: Dec 30 10:26:13 2018 GMT
*        expire date: Feb 28 16:23:00 2021 GMT
*        subjectAltName: rapi.recombee.com matched
*        issuer: C=US; ST=Arizona; L=Scottsdale; O=GoDaddy.com, Inc.; OU=http://certs.godaddy.com/repository/; CN=Go Daddy Secure Certificate Authority - G2
*        SSL certificate verify ok.
> POST /telexa-dev/batch/?hmac_timestamp=1585732597&hmac_sign=XXXHTTP/1.1
Host: rapi.recombee.com
Accept: */*
Accept-Encoding: deflate, gzip
Referer: https://rapi.recombee.com/telexa-dev/batch/?hmac_timestamp=1585732597&hmac_sign=XXX
Content-Type: application/json
User-Agent: recombee-php-api-client/3.0.0
Connection: close
Content-Length: 658192
Expect: 100-continue

[OndraFiedler]

Thanks for opening the ticket.
Support for GZIP definitely makes sense, however I assume it is not the main bottleneck as the requests & responses are rather small.

The ways for achieving higher throughput will be:

1. Assigning more computational resources for your database
2. Moving your database to a datacenter near you - I assume to North America? Currently the databases are created in Europe by default.

Both can by done by Recombee support, so please contact [email protected].

[staabm]

please see https://blackfire.io/profiles/6d579fdc-bb0a-4c92-aebe-bde02484ee61/graph for a in-depth look into our cronjob.

the curl requests take the most of the time. not sure whether the time is spent in-transit or at your computational end.

I will contact the suppport for now and see what I can achieve.

thx for the fast feedback

[staabm]

Support for GZIP definitely makes sense, however I assume it is not the main bottleneck as the requests & responses are rather small.

I guess in our case, where we send 10.000 items for deactivation having them GZIP'ed (or at least somehow compressed) could make a big difference.

as you can see in the above blackfire profile, we spent ~40 seconds to delete 10.000 items:

[staabm]

just got some numbers from production. there we have 4x ~40 seconds delete commands for 10.000 item batches each

[staabm]

it seems we are again running into this problem.

Support for GZIP definitely makes sense, however I assume it is not the main bottleneck as the requests & responses are rather small.

was GZIP support activated on your end since then?

[Pazekal90]

it seems we are again running into this problem.

Support for GZIP definitely makes sense, however I assume it is not the main bottleneck as the requests & responses are rather small.

was GZIP support activated on your end since then?

Ping @OndraFiedler. Is there something new?